A preliminary model of Ikpana intonational phonology

Hironori Katsuda

1 Introduction

Ikpana is an endangered language spoken by approximately 7,500 people in south- eastern . It belongs to the Kwa branch of the Niger-Congo family, and is also counted as one of the 15 Ghana- Mountain (GTM) languages. Although its basic phonology and morphosyntax have been documented by Dorvlo (2008), little is known about its suprasegmental properties. Accordingly, this study proposes a preliminary model of the intonational phonology of

Ikpana within the autosegmental-metrical (AM) framework of intonational phonology (Beckman

& Pierrehumbert, 1986; Ladd, 1996/2008; Pierrehumbert, 1980), expanding empirical knowledge of intonation in African tone languages, and contributing to understanding of prosodic typology, especially with reference to intonationally-defined prosodic units (e.g., Beckman & Pierrehumbert,

1986; Jun; 1998; Pierrehumbert & Beckman, 1988) and question prosody (Clements & Rialland,

2008; Rialland, 2007; 2009).

1.1 Intonation of African tone languages

Tone languages manipulate F0 at the lexical level to distinguish words and certain grammatical functions. Furthermore, it is well established that tone languages, as well as non-tone languages, use F0 variation at the postlexical level to convey a certain pragmatic meaning and sentence type information (Ladd, 1996/2008). However, due to the high functional load of F0 at the lexical level, tone languages are assumed to have limited use of F0 at the postlexical level

(Hyman & Monaka, 2011). The literature has explored how tone languages encode the information

1

that is typically expressed by postlexical F0 variation in non-tone languages without obscuring or changing lexical or grammatical meaning (e.g., Cruttenden, 1997; Hyman & Monaka, 2011; Yip,

2002).

Intonation studies of African tone languages have attempted to distinguish a postlexical use of F0 variation from a lexical use. In her pioneering work, Lindau (1986) tested a formal model of the relation between tone and intonation in Hausa, a two-tone Chadic language spoken in Nigeria, by investigating sentences with different tone patterns in several sentence types. She found that the degree of downward F0 slope of an utterance varies depending on several factors, such as sentence type, sentence length, and tone pattern. For example, like-tone statements are characterized by a downward F0 slope, while like-tone questions (yes-no questions and wh- questions) are produced with zero slope. Furthermore, there are some intonational differences between yes-no questions and wh-questions: analyses of alternating tone sentences reveal that wh- questions are more susceptible to downdrift (i.e., an F0 lowering of a H tone after a L tone) than yes-no questions, and only yes-no questions are characterized by an F0 rise of an utterance-final

H tone, which is sometimes followed by an F0 fall.

Numerous studies that follow Lindau (1986) have also determined how African tone languages can use phrasal level F0 manipulation. In the Kwa branch of the Niger-Congo family, the most well studied language in terms of intonation is probably Akan, which is a two-tone language spoken in Ghana and some eastern parts of the . According to Genzel (2013) and Kügler (2017), simple statements in Akan are marked by a neutralization of the two lexical tones (H and L) in the sentence-final position, which is caused by final lowering of a H tone.

Moreover, yes-no questions are marked by a variety of prosodic markers: L% boundary tone,

2

lengthening of sentence-final vowel, a higher intensity as well as breathy termination, and a raised pitch register. Wh-questions, however, exhibit only a pitch register raising as a prosodic marker.

As shown in the cases of Hausa and Akan, intonation can interact with lexical tones in several ways. One of the central issues in intonation studies of African tone languages has been to typologize the ways intonation interacts with lexical tones. Hyman and Monaka (2011) suggest that languages with a complex tonal system may exhibit no boundary tones, and this seems to be consistent with the observations that Yoruba, a three-tone language (Connell & Ladd, 1990;

Laniran, 1992), and Mambila, a four-tone language (Connell, 2017), do not use boundary tones.

Hyman and Monaka propose three possible interaction patterns between lexical tones and boundary tones: ‘accommodation’, ‘submission’, and ‘avoidance’. In accommodation, both tone and intonation are realized but on different syllables, or intonation superimposes tone to give different pitch levels or pitch ranges. In submission, lexical tone is overridden by intonation, while in avoidance, intonation is minimized.

As is true in both Hausa and Akan, African tone languages are also known to have diverse question prosody markers. Rialland (2007; 2009) (see also Clements & Rialland, 2008) investigated yes-no question markers of more than 70 African languages and found that almost half of them display some type of non-high-pitched markers (collectively referred to as ‘lax’ question prosody), as opposed to high-pitched markers (collectively referred to as ‘tense’ question prosody), which had hitherto been considered as (near-) universal (Bolinger, 1978; Ohala, 1984).

The inventory of each category is listed as follows:

(1) Lax yes-no question prosody a. final L tone or final intonation (L%) b. final lengthening 3

c. breathy termination d. [open] vowel

(2) Tense yes-no question prosody a. cancellation/reduction of downdrift, register expansion b. raising of last H(s) (not necessarily sentence-final) c. cancellation/reduction of final lowering d. final H tone or rising intonation (H%) e. final HL melody

A language generally employs more than one marker to characterize its question prosody, and some languages have more than one question prosody, each of which has its own combination of markers. Furthermore, it is possible for a language to use markers from both sets for its question prosody. This is called ‘hybrid’ question prosody and describes cases such as those in Akan: (a)

L% boundary tone, (b) lengthening of sentence-final vowel, and (c) breathy termination in the lax prosody inventory (1), while (a) a raised pitch register in the tense prosody inventory (2).

Crucially, the Kwa family, together with the Gur and Kru families, constitutes the ‘core’ area where lax prosody markers are used. Nine out of ten , including Ewe (Westerman,

1930) and Akan (Genzel, 2013; Kügler, 2017), use a L% boundary tone (or a final L tone) to mark a polar question. The only Kwa language that is not reported to show lax question prosody is Ga, in which (a) cancellation/reduction of downdrift, register expansion, and (c) cancellation/reduction of final lowering in the tense prosody inventory (2), are used to mark yes-no questions (Rialland,

2007). Other than languages included in Rialland’s database, Cahill (2015) reports that two Kwa languages, Adele and Chumburung, and three , Buli, Deg, and Safaliba (all five languages are spoken in Ghana), all exhibit some sort of falling pitch in yes-no questions. However,

4

recent work by Essegbey (in press) reports that Tutrugbu (also known as Nyangbo), a Kwa language, exhibits a slight F0 rise at the end of yes-no questions. In fact, Tutrugbu is more closely related to Ikpana than other Kwa languages, since Tutrugbu is also counted as one of the GTM languages.

This study extends our knowledge of possible prosodic structure in African tone languages, and prosodic typology more generally. More specifically, this study shows that Ikpana has a robust prosodic system with three levels of prosodic units, which is comparable to those of non-tone languages. This finding is consistent with the claim the intonational features of a language are not directly predictable from its lexical prosody (Jun, 2005).

1.2 The present study

The aim of this study is to propose a preliminary model of the intonational phonology of

Ikpana, a tone language spoken in south-eastern Ghana, expanding empirical knowledge of intonation in African tone languages, and contributing to understanding of prosodic typology. The theory of intonation adopted in this paper is the AM framework of intonational phonology

(Beckman & Pierrehumbert, 1986; Ladd, 1996/2008; Pierrehumbert, 1980). Although prosodic systems of African tone languages have often been analyzed using the so-called syntactic approach, such as the prosodic phonology or the indirect syntactic approach (Nespor & Vogel, 1986; Selkirk,

1984; 1986), in which prosodic units are derived indirectly from the syntactic structure, adopting the AM framework has the following advantages. First, it allows for direct comparison with intonational systems that have different types of lexical prosody, such as lexical stress and lexical pitch accent (Jun, 2005; 2014). Second, it can capture the cases where prosodic phrasing cannot be determined purely based on the syntactic structure (e.g., Jun, 1998 on Korean accentual phrase).

5

There is a case in Ikpana where a prosodic unit, which I refer to as an intermediate phrase, does not correspond to the phonological phrase in the syntactic approach. For these reasons, I adopt the

AM framework, in which prosodic units are intonationally defined based on the surface phonetic form of an utterance. The analyses presented in this paper suggest that Ikpana has a robust intonational system with three prosodic units, and that its intonation can be modeled in the same way as that of non-tone languages.

The remainder of the present paper is structured as follows. A brief introduction to Ikpana, including the tonology of the language, will be presented in Section 2. Section 3 introduces a preliminary model of Ikpana intonational phonology based on declarative sentences. In Section 4,

I will discuss the intonation of other sentence types―yes-no questions, wh-questions, and imperatives―and compare it with the intonation of declarative sentences. Finally, Section 5 summarizes the proposed model of Ikpana intonational phonology.

2 Background

2.1 Ikpana

Ikpana (also known as Logba) is an endangered tone language spoken by approximately 7,500 people in the Logba region of Ghana (Dorvlo, 2008). It belongs to the Kwa branch of the Niger-

Congo family and is also counted as one of the 15 GTM languages spoken in the Akuapem-Togo-

Atakora hills ranging across the Ghana-Togo border (Ameka, 2017; Ameka & Essegbey, 2017).

The GTM languages are known for their uncertain phylogenetic situation. Although researchers have attempted to classify them based on their specific linguistic characteristics, no firm conclusion has been reached (Blench, 2006; 2009). Among the GTM languages, Ikpana was one of the least studied when Dorvlo (2008) wrote his dissertation on its grammar.

6

The Logba region contains the following eight villages: Vuinta, Ogɔme, Akusame, Adiveme,

Adzakoe, Alakpeti, Tota, and Klikpo. Although its dialectal variation is not well understood, native speakers tend to acknowledge that the Tota dialect, which is spoken in the hills, is distinct from the other dialects, particularly that spoken in Alakpeti, the commercial center of Logba (Dorvlo,

2008). The data analyzed in this study were elicited from one Tota dialect speaker and two Alakpeti dialect speakers. However, since no dialectal differences were found in their supra-segmental properties, there will be no description of the individual dialects.

2.2 Brief outline of the linguistic structure of Ikpana

2.2.1 Basic features of Ikpana morphosyntax

Ikpana has an SVO word order with noun modifiers, such as adjectives and demonstratives, following the head noun. Unlike other Kwa languages, which are known to lack a noun class system, Ikpana, along with the other GTM languages, has a noun class system resembling that of the Bantu languages, where nouns are prefixed by a class marker based on semantic characteristics.

According to Dorvlo (2008), Ikpana class markers include vowel prefixes (a-, i-, u-, e-/ɛ-, o-/ɔ-) and a nasal prefix (n-). Moreover, the verb roots take a subject agreement marker as a prefix, which agrees with the subject in class. A tense/aspect marker can be placed between the subject and the verb stem, although the distinction between the present and past tenses is expressed only via tones.

2.2.2 Basic features of Ikpana segmental phonology

Ikpana has three allowable syllable types: vowel or only (V/N), onset plus nucleus (CV), and onset cluster plus nucleus (CCV). According to Dorvlo’s (2008) analysis,

Ikpana has seven vowels (/i, e, ɛ, a, u, o, ɔ/) with an advanced tongue root (ATR)

7

system such that the ATR value of the stem controls that of the prefixes. Only the mid vowels (/e,

ɛ, o, ɔ/) have a [± ATR] distinction, while /a/ and the two high vowels (/i, u/) have no harmonic partners.

Since nouns and verbs often have a vowel prefix (i.e., a class marker for nouns and a subject marker for verbs), while coda consonants are not allowed, vowel hiatus abounds in Ikpana sentences. As a strategy to resolve this hiatus across a word boundary, the vowel before the word boundary (V1), as opposed to the vowel after the boundary (V2), is often deleted or reduced. In my data, deletion is more common and happens in a wide variety of contexts, including cases where the adjacent vowels are identical. For example, the sentence uma ɔri imunyi ‘The mother holds hair’ is realized as um[ɔ] r[i] munyi in natural speech, which is shown in Figure 1.

Figure 1. Sample pitch track showing vowel deletion and subsequent re-syllabification

Here, the final vowel of uma (‘the mother’) and the vowel of the verb stem ri (‘hold’) are deleted and not shown in the Ikpana words tier (the second tier from the top). I analyze the subject marker

/ɔ/ as being re-syllabified with the consonant /m/ of uma (shown as um[ɔ]), while the class marker

/i/ of the following noun imunyi (‘hair’) as being re-syllabified with the verb onset /r/ (shown as r[i]). Although all tones are marked as H on the lexical tone tier (the top tier), the tone on the verb

8

ri is higher than the rest of the lexical H tones, which is denoted by an upward arrow after the lexical tone. This phenomenon will be discussed in detail in 3.2.1.

On the other hand, reduction often happens when V1 is /i/ and V2 is different from /i/. I analyze this reduction as an incomplete realization of deletion, as the phonetic realization of the reduced vowel is often not short enough to be a glide, which is a common resolution strategy cross- linguistically (Casali, 1996). Figure 2 shows a sample pitch track including a reduced vowel, in which the vowel of the verb ri (‘hold’) is reduced, which is marked by the extra-short diacritic mark (˘) over the vowel (i.e., rĭ), followed by the class marker /a/ of the following noun azayi

(‘beans’).

Figure 2. Sample pitch track showing a vowel reduction signaled as rĭ.

As in Figure 1, the pitch on the H-tone verb ri is higher than the preceding H tones, and the following L tone on the first syllable of azayi is also raised. Again, this will be discussed in 3.2.1.

Although deletion or reduction of V1, while preserving V2, is more common in Ikpana and is consistent with the cross-linguistic tendencies found in Casali (1997), deletion of V2 is also possible in certain contexts, especially when V1 is a part of a grammatical word. In Figure 3, the vowel hiatus between kpɛ (‘and’) and utsi (‘the father’) are resolved by the deletion of the class marker of utsi (i.e., V2), which is not shown in the Ikpana words tier. 9

Figure 3. Sample pitch track showing deletion of V2. The coordinator kpɛ is tonally unmarked

(marked by ∅ in the figure) in the Dorvlo’s (2008) dictionary

Furthermore, the vowel hiatus resolution between a subject and a verb exhibits a different pattern compared to that formed in other locations (e.g., between a verb and an object). Between a subject and a verb, the final vowel of the subject is deleted as in other locations, but the compensatory lengthening of the subject marker vowel /ɔ/ also occurs. Moreover, when the subject is a L-tone noun, the H tone of the subject marker is usually realized toward the end of the lengthened vowel. In Figure 4, the sentence ina ɔnɛ azayi is pronounced as in[ɔ] n[a] zayi in natural speech. The final vowel of the subject ina (‘a person’) is deleted, and the following subject marker vowel /ɔ/ is re-syllabified with the consonant /n/ of the subject, showing a compensatory lengthening and the subject marker H tone. This particular pattern seems to be related to the presence of the subject marker, as it is the only vowel hiatus that involves the subject marker and other vowel hiatuses involve a class marker. However, the compensatory lengthening and the subject marker H tone are sometimes absent when the sentence is longer. An example of this is shown in Figure 18 in Section 3.2.3.1.

10

Figure 4. Sample pitch track showing a compensatory lengthening and the subject marker H tone at the vowel hiatus between a subject and a verb.

In numerous languages, hiatus resolution is conditioned by prosodic domains (Mudzingwa,

2010). For example, it does not occur across an intermediate phrase boundary in Greek (Arvaniti

& Baltazani, 2005; Baltazani, 2006) and Uyghur (Major & Mayer, 2018). In African languages,

Emai (Volta-Niger, Nigeria) only allows vowel deletion within lexical phrases, such as NP and

VP (Egbokhare, 1990), while Yoruba allows it to occur at the phrasal level (e.g., between a verb and its object, between a preposition and its object, and between a conjunction and the following conjunct) (Pulleyblank, 1988). In this respect, Ikpana is more similar to Yoruba, allowing vowel hiatus resolution to occur at the phrasal level. I will describe it more detail in Section 3.2.3.3 and argue that vowel hiatus resolution in Ikpana is conditioned by the presence or absence of an intervening Intonational Phrase boundary.

11

2.3 Basic features of Ikpana tonology

2.3.1 Lexical tone

Ikpana is a tone language where differences in pitch indicate lexical differences in meaning.

Dorvlo (2008) identifies two basic tones, namely High (H) and Low (L), and several tonal minimal pairs, such as lá (‘make, prepare’) and là (‘beat’). Moreover, although he does not mention it explicitly, he identifies the existence of tonally unmarked syllables. In the dictionary section of his dissertation, these syllables are unmarked as opposed to being marked with an acute accent, to denote the H tone (e.g., á), or a grave accent, to denote the L tone (e.g., à). The tonally unmarked syllables often appear as noun class markers, but sometimes appear in roots. Conversely, there are also some nouns that have either a H-tone or a L-tone class marker. Since the status of the tonally- unmarked syllables is still unclear, this study focuses only on H and L tones.

2.3.2 Grammatical tone

In addition to its lexical functions, the tone can serve a grammatical function. One example of this is the tone. Although Dorvlo (2008) does not identify a clear-cut distinction between the past and present tense in Ikpana, I observe that the distinction is signaled by tones on the subject marker as well as on the verb stem.

Figure 5 presents the F0 values of H- and L-tone verb stems (V), as well as the preceding subject marker (SM) split by the present and past tenses. In the present tense, the subject marker is realized with a H tone, regardless of the lexical tone of the following verb stem, while the verb stems analyzed in Dorvlo (2008) that bear a lexical H tone (ri ‘hold’) and a lexical L tone (nɛ

‘buy’), respectively, are realized with a pitch one step higher than its expected level. More specifically, the H-tone verb is realized with even higher pitch than the H tone level, a pitch level

12

that I will refer to as extra-high, while L-tone verbs are realized as high as H tones. Since the extra- high pitch level is generally limited to a word-final position, this study treats it as phonetic pitch realization instead of underlying tonal category. In Section 3.2.1, I will argue that this word-final pitch raising is not restricted to verbs but also happens to other lexical categories, as long as the word is not in a phrase-final position. In the past tense, on the other hand, the subject marker is marked by a lower pitch, and both H- and L-tone verb stems are realized even higher than the corresponding verb stems in the present tense.

Figure 5. The average F0 values of H- (red color) and L-tone (blue color) verb stems, as well as the preceding SM, split by the present (shown by a filled shape connected by a solid line) and past tenses (an empty shape connected by a dotted line). Each data point represents the average F0 values of two different sentences (uma ɔri imunyi ‘The mother holds/held hair’, and ina ɔri imunyi ‘A person holds/held hair’ for the H-tone verb stem ri (‘hold’); uma ɔnɛ imunyi ‘The mother buys/bought hair’, and ina onɛ azayi ‘A person buys/bought beans’ for the L-tone verb stem nɛ (‘buy’) read by one male speaker.

13

Overall, the past tense makes the tone on the subject marker lower and that on the verb stem higher than the corresponding verb stem in the present tense. The data analyzed in the later sections consist only of present tense sentences, as their less extreme tonal realization is more appropriate for observing the effect of F0 at the postlexical level.

3 Intonational phonology of Ikpana

In this section, I will present a preliminary model of Ikpana intonational phonology within the AM intonation framework. I will begin by describing the data collected through fieldwork in

Ghana, in Section 3.1. Section 3.2 outlines the basic prosodic structure of Ikpana, which I propose involves three prosodic units at and above the word level (namely the prosodic word (PW), the intermediate phrase (ip), and the Intonational Phrase (IP)), and discusses how each prosodic unit is marked prosodically. Section 3.3 describes the intonation of other sentence types, including yes- no questions, wh-questions, and imperatives.

3.1 Data and methodological procedures

The data analyzed in this study were elicited from three native speakers (two males and one female: M1 (70s, Tota), M2 (70s, Alakpeti), and F1 (50s, Alakpeti), during a six-week fieldwork trip to Ghana, conducted in the summer of 2018.

To distinguish intonational pitch variation from lexical tone, I constructed tonally controlled sentences where all but a few exceptional syllables have the same lexical tones (i.e., all H tones or all L tones), or where H- and L-tone syllables are mixed on purpose (e.g., alternating tones:

HLHL…) (Lindau, 1986; Connell & Ladd, 1990). The types of sentences elicited were as follows:

14

(3) a. 48 like-tone simple (mono-clausal) declaratives (28 H-tone and 20 L-tone sentences) and the corresponding yes-no questions b. 62 mixed-tone declaratives and the corresponding yes-no questions c. 26 like-tone complex (bi-clausal) declaratives (13 H-tone and 13 L-tone sentences) and the corresponding yes-no questions of 10 of the 26 complex declaratives (plus 14 like-tone complex sentences from M1) d. 28 wh-question and answer pairs (plus 13 wh-question and answer pairs from M1) e. eight imperatives

As mentioned earlier, all the sentences are in the present tense to avoid introducing pitch raising on the verb stem. Their grammaticality was also confirmed by the native speakers.

Moreover, the sentences were varied in length and syntactic structure (due to e.g., the presence or absence of an adjective and a postpositional phrase) to investigate whether these factors affect prosodic phrasing.

The sentences were randomized and presented in both Ikpana and English on a computer screen, one after another. The English versions of the sentences were provided to clarify any ambiguous meaning, since there are several tonal minimal pairs in Ikpana. The speakers were instructed to produce each sentence in two speech styles, namely careful speech and natural speech, with a neutral focus. Each sentence was repeated three times.

The recordings were analyzed in Praat (Boersma & Weenink, 2017) with four tiers: lexical tones, Ikpana words, the English gloss of each Ikpana word, and the English translation of the sentence. In the Ikpana words tier, the deleted vowel is not shown, the re-syllabified vowel is enclosed in square brackets, and the original position of the re-syllabified vowel is enclosed in parentheses. Furthermore, vowel reduction is marked by the extra-short diacritic mark (˘) over the vowel.

15

3.2 Prosodic structure

Analysis of the data described above reveals three prosodic units at or above the word level marked by intonation, namely the PW, the ip, and the IP, defined by the edge tones and several other prosodic cues associated with each prosodic unit. The latter include pitch reset, pause, and final lengthening. Furthermore, vowel hiatus resolution was also found to be conditioned by the presence or absence of an intervening IP boundary, and thus serves as one of the diagnostics that can be used to distinguish ip from IP boundaries.

3.2.1 Prosodic word (PW)

Among the three prosodic units in Ikpana, the smallest is the PW, which is posited as a lower constituent than the Accentual Phrase in the prosodic hierarchy. The PW consists of a root or stem and affixes, while the Accentual Phrase can contain one or more PWs (Jun, 1993). As in the case of the Accentual Phrase, the PW is argued to be marked by intonational tones in several languages, such as Turkish (Ipec, 2015), and Serbo-Croatian (Godjevac, 2005).

Evidence of the PW in Ikpana is an F0 rise, marking the right edge of a PW in both H-tone and L-tone sequences. This is analyzed as a High boundary tone marking the right edge of a PW, labelled as Hω. A similar phenomenon is observed in Baule (Kwa, the Ivory Coast) (Leben &

Ahoua, 1997), where sequences of H tones are marked by a gradual F0 rise over the PW. The authors call this ‘upsweep’ and refer to the F0 peak at the right edge as Super-High. Furthermore, they also note that there is an alternative variant of upsweep, where only the final H tone in the

PW is raised, while the preceding tones remain flat. The Hɷ in Ikpana is similar to this variant in

Baule, raising only the right-edge tone.

Figure 6 shows four instances of the Hɷ tones, marking the right edge of the verb r[i] (‘hold’),

16

the object noun imunyi (‘hair’), and both the noun and adjective of the noun phrase ɔyɔ bibi (‘a small tree’), which is the complement of the postposition otsu (‘on top of’). At the hiatus between ri and imunyi, the vowel of ri is deleted, resulting in the class marker of imunyi being re-syllabified as the nucleus of the verb (i.e., r[i]) and realized with the extra-high pitch of the verb-final tone.

At the hiatus between bibi and otsu, on the other hand, the final vowel of bibi is reduced, and the extra-high pitch of the vowel is extended to the first vowel of the following postposition.

Furthermore, there are some other lexical or postlexical processes observed in this pitch track:

(i) the first syllable of both the first and second adjectives (both are bibĭ (‘a little, small’)) is raised in pitch, (ii) the second syllable of the first adjective is lowered in pitch, and (iii) the second syllable of the postposition is lowered in pitch. (i) is discussed later in this section, and (ii) and (iii) are examined in Sections 3.2.2 and 3.2.3.1 respectively, in detail.

Figure 6. Sample pitch track showing PW-boundary tones (Hɷ), marked by upward arrows.

Figure 7 (left) shows the average F0 values on each syllable of a H-tone noun imunyi (‘hair’), which was produced after the H-tone monosyllabic verb ri (‘hold’) or disyllabic verb ŋɔnyi (‘take something by force’), and always followed by bibi (‘little’) in the data. Solid lines represent F0 values in natural speech, while dotted lines represent those in careful speech. Different colors are

17

used for different speakers. In natural speech, the first vowel of the noun is missing, since the first syllable of the noun is re-syllabified as a nucleus of the preceding syllable (i.e., r[i]) in natural speech, it is not shown in the figure. A mixed-effect linear regression with individual speaker as the random intercept indicates that the F0 values of the final syllable of the noun are higher than those of the second syllable in both natural and careful speech (p > .001), while the F0 values of the first two syllables in careful speech are not significantly different (p = .623).

Figure 7. Left: Average F0 values on each syllable of the H-tone noun imunyi (‘hair’). (solid line: natural speech, dotted line: careful speech). The values are based on three repetitions of 12 different all-H sentences. Right: Average F0 values on each syllable of a L-tone noun azayi (‘beans’). The values are based on the three repetitions of eight different all-L sentences.

Figure 7 (right) shows the average F0 values of a L-tone noun azayi (‘beans’), which is an object of the L-tone verb nɛ (‘buy’) or lili (‘grate’), and always followed by vuvɔ (‘new’) in the data. Again, the first syllable of the noun is missing in natural speech because it is re-syllabified as the nucleus of the preceding syllable due to vowel hiatus resolution. As in the H-tone noun imunyi, the F0 values of the final syllable are significantly higher than those of the second syllable

(p < .001). Unlike the H-tone noun, however, the F0 values of the first syllable are significantly 18

higher than those of the second syllable (p > .001). Since this happens in the careful speech, where vowel hiatus resolution does not occur, I argue that, in Ikpana, H-tone sequences tend to maintain flat F0 values over the first and the second syllable of the noun, while L-tone sequences tend to fall gradually; this has also been observed in other languages such as Yoruba (Connell & Ladd,

1990; Laniran, 1992), Mambila (Connell, 2003), and Ibibio (Urua, 2002).

The Hɷ is also observed in adjectives, provided that the adjective is not in a phrase-final position. Figure 8 (left) shows the average F0 values of a H-tone phrase consisting of ɔyɔ (‘a tree’) and bibi (‘small’), which is followed by H-tone postposition etsi (‘under’) or otsu (‘on top of’) to form a postpositional phrase (PP) ɔyɔ bibi etsi (‘under a small tree’) or ɔyɔ bibi otsu (‘on top of a small tree’). It appears that all speakers exhibit an F0 rise at the end of the noun ɔyɔ, while only the male speakers (M1 and M2) exhibit an F0 rise at the end of the adjective bibi, except for the final syllable in M1’s natural speech, which is just as high as the preceding syllable; this is likely due to a tonal assimilation to the first syllable of the following postposition. It should also be noted that the first syllable of the adjective bibi remains relatively high (it is slightly lower than the second syllable of the noun ɔyɔ in the male speakers’ speech, while higher in F1’s speech). While it is probable that the first syllable of the H-tone adjective bibi is realized as an extra-high pitch, it is not clear whether this pitch raising is caused by a noun-adjective combination or is simply a pitch realization pattern of adjectives. Also, since bibi is the only HH-tone adjective I tested, it is also possible that the extra-high pitch realization of the first syllable is an idiosyncratic feature of the specific adjective1. It seems that F1’s F0 lowering on the adjective is also due to a tonal assimilation of the final tone of the adjective to the first tone of the following postposition etsi

1 Dorvlo (2008) analyzes that bibi is derived from a noun via .

19

(‘under’) or otsu (‘on top of’). Since the second vowel of bibi is reduced at the vowel hiatus, the influence of the following tone can be greater. It is necessary to check more H-tone adjectives to ascertain if the Hɷ is consistent in adjectives.

Figure 8. Left: Average F0 values of a H-tone phrase ɔyɔ bibi (‘a little hair’). The values are based on three repetitions of six all-H sentences. Right: Average F0 values of a L-tone phrase ɔyɔ vuvɔ (‘new beans’). The values are based on three repetitions of four all-L sentences.

On the other hand, the Hɷ is clearer in a L-tone phrase consisting of ɔyɔ (‘a grave’) and vuvɔ

(‘new’) when it forms a PP ɔyɔ vuvɔ nu (‘in a new grave’). As is clear from Figure 8 (right), the second syllable of each word is raised. This time, the first syllable of the L-tone adjective vuvɔ is even lower than the first syllable of the noun, reflecting a greater effect of declination on L-tone sequences.

Although quantitative data are not shown here, verbs also show the Hɷ. As shown in Figure

1, H- and L-tone stem’s final syllables are realized with a pitch one step higher than the expected pitch level, and this is also true in disyllabic H-tone verb stems, such as ŋɔnyi (‘take something by force’), as shown in Figure 9. However, disyllabic L-tone verb stems, such as lili (‘grate’) and

20

ŋɔnyi (‘write’) do not exhibit the Hɷ. Since disyllabic L-tone verb stems are extremely rare in

Dorvlo’s (2008) dictionary, verbs with this tonal pattern can be recognized somehow differently from other lexical items. Thus, I treat them as an exception to the Hɷ for now.

Figure 9. Sample pitch track showing the Hɷ at the end of a H-tone disyllabic verb stem, marked by an upward arrow.

Overall, it is found that the Ikpana PW is marked by an F0 rise on its right edge (Hɷ), which is observed in object nouns, adjectives, and verbs (except for disyllabic L-tone verb stems).

However, subject nouns are not consistent in terms of the Hɷ. The noun utsi (‘the father’), which consists of two H tones, appears to exhibit the Hɷ in the subject position, but since the consonant before the second vowel is an obstruent, and the second vowel is usually reduced at vowel hiatus

(i.e., utsĭ), it is hard to tell whether it is due to micro-prosody or the Hɷ. On the other hand, the noun uma (‘the mother’) does not exhibit the Hɷ at all. It is also possible that there is a bigger prosodic boundary, which I will discuss in the next section, between a subject and a verb, and it prevents the final syllable of the subject from being raised. However, since there are only two nouns (uma and utsi) used as a subject in my corpus, it is not clear whether the lack of the Hɷ in

21

‘the mother’ is due to the bigger boundary or an idiosyncratic feature of the specific noun. More nouns need to be examined in the subject position to decide this.

3.2.2 intermediate phrase (ip)

Cross-linguistically, the ip (or phonological phrase in other analyses) typically aligns with a syntactic constituent (Truckenbrodt, 1999). In Ikpana, it generally corresponds to a lexical maximal projection (XP), and is marked by phrase-final lowering, for which I propose a L boundary tone, labelled L-. In Figure 10, ip-final lowering is observed on the last syllable of the first instance of bibi (‘a little’), which appears at the right edge of the object NP imunyi bibi (‘a little hair’), and the pitch range is reset at the beginning of the following PP. It should be noted that the second instance of bibi (‘small’) located in the middle of the PP does not show final lowering. The difference between the two instances of bibi suggests that the Hɷ of a PW boundary is overridden by the L- in the ip-final position. Final lowering is also observed on the sentence- final syllable; however it is analysed here as a L% Intonational Phrase (IP) boundary tone, whose lowering effect often extends to more than one syllable, and usually exhibits a greater degree of lowering than an ip-final L- boundary tone, as it marks the sentence-final IP boundary. I will discuss this in detail in the next section.

Figure 10. Sample pitch track showing ip-final lowering (L-), marked by a downward arrow.

22

Figure 11. Sample pitch track showing a lack of L-, marked by an upward arrow, when the object NP consists only of a noun.

However, syntactic constituency does not suffice to capture the environment in which the L- occurs. For example, the juncture between an object NP and the PP is not marked by an ip boundary when the syntactic constituents are short, e.g., with only a one-word object NP followed by a two- word PP, as in Figure 11. This suggests that phonological factors also play a role in determining the ip domain.

In fact, the juncture between a subject and the following verb is marked by an ip-boundary when the subject consists of a noun and an adjective, as in Figure 12, or is even longer, involving two NPs coordinated by the coordinator kpɛ ‘and’, as in Figure 13. In both cases, the final syllable of the subject NP, which is the second syllable of the adjective bibi, is lowered due to the L- boundary tone. It should be noted that the second syllable of the first instance of bibi in Figure 13 does not show the L-, since that word is located in the middle of the subject NP.

23

Figure 12. Sample pitch track showing ip-final lowering (L-) at the end of a subject consisting of a noun and an adjective, marked by a downward arrow.

Figure 13. Sample pitch track showing ip-final lowering (L-) at the end of a subject consisting of two NPs coordinated by kpɛ ‘and’, marked by a downward arrow. The coordinator kpɛ is tonally unmarked (marked by ∅ in the figure) in the Dorvlo’s (2008) dictionary.

3.2.3 Intonational Phrase (IP)

3.2.3.1 Pause and boundary tones: L% and H%

The IP is the largest prosodic unit in Ikpana, and is marked by a boundary tone, which is either L% or H%, followed by a pause. A sentence-final IP is usually marked by L% in declaratives and lowers the final tone of the sentence. Crucially, the visibility of the effect of a L% boundary tone depends on the tonal context of the final and penultimate tones. The sentences in Figures 14 24

and 15 have an extra-high pitch on the penultimate syllable. In Figure 14, the extra-high realization of the verb ri extends to the following syllable due to the vowel reduction of the verb-stem vowel, i.e., rĭ. In Figure 15, the first syllable of the H-tone adjective bibi is realized with an extra-high pitch. When the penultimate syllable is realized with an extra-high pitch, the falling movement toward the L% target is clear.

Figure 14. Sample pitch track showing L% when the penultimate syllable is realized as extra-high (since the extra-high pitch on the verb extends to the penultimate syllable, due to vowel reduction) and the final syllable bears a lexical H tone.

Figure 15. Sample pitch track showing L% when the penultimate syllable is realized as extra-high (since the first syllable of bibi ‘small’ is the extra-high pitch) and the final syllable bears a lexical H tone.

25

On the other hand, the sentences in Figures 16 and 17 have a H pitch on the penultimate syllable.

In Figure 16, the vowel of the L-tone verb nɛ is deleted due to the vowel hiatus resolution, and the initial vowel of the noun ɔyɔ bears the verb L-tone, which is realized as a H pitch. In Figure 17, the sentences end with the noun imunyi, the penultimate syllable of which is a H pitch. In this case, the final tone is only slightly lower than the penultimate tone. In fact, the average F0 values of the sentence-final syllable of Figure 14: uma ɔri ɔyɔ ‘The mother holds a tree’, which is pronounced as um[ɔ] rĭ ɔyɔ and Figure 16: uma ɔnɛ ɔyɔ ‘The mother buys a tree’, which is pronounced as um[ɔ] n[ɔ] yɔ by Speaker M1 are quite similar: 109 Hz for the former and 113 Hz for the latter.

So, one can say there is no L% on the sentence-final syllable in these sentences, but instead, the seemingly low F0 on the final syllable is a by product of the extra-high on the penultimate syllable.

That is, the final syllable’s F0 is still high, but looks low compared to the preceding extra-high tone. However, as shown in Figures 10 and 13, F0 after extra-high is not always lowered. The lowering of H after extra-high happens at the boundary of a heavy syntactic constituent.

Figure 16. Sample pitch track showing L% when the penultimate syllable is realized as H (since the verb L tone, which is realized as a H pitch, is realized on the penultimate syllable, due to vowel deletion) and the final syllable bears a lexical H tone.

26

Figure 17. Sample pitch track showing L% when the penultimate syllable is realized as H (since the penultimate syllable of the noun imunyi (‘hair’) is a H pitch) and the final syllable bears a lexical H tone.

The same relationship between the pitch height of the penultimate syllable and the visibility of L% holds in L-tone sentences too. Figure 18 shows the pitch track of a L-tone sentence where the penultimate syllable is raised due to the Hω marking the right edge of the L-tone word ɔyɔ (‘a grave’). Following the raised syllable, the final lexical L tone is lowered due to the effect of L%.

Figure 18. Sample pitch track showing L% when the underlying L tone on the penultimate syllable is realized as H tone at a phrasal level due to Hω, and the final syllable bears a lexical L tone. In this sentence, the compensatory lengthening and the subject marker H tone are absent (shown as H? in the lexical tone tier).

27

In Figure 19, on the other hand, the penultimate syllable of the L-tone sentence is not raised, and the final syllable is only slightly lower than the penultimate syllable.

Figure 19. Sample pitch track when the penultimate syllable is realized as L pitch and the final syllable bears a lexical L tone.

One of the important differences between the L- marking the ip and the L% marking the IP is that the L% does not have the weight constraint that the L- has. That is, a L% can occur without having enough phonological weight that a L- requires to occur. For example, the sentence in Figure

11, um[ɔ] r[i] munyĭ ɔyɔ̆ etsi (uma ɔri imunyi ɔyɔ etsi) ‘The mother holds hair under a tree’ does not show a L- between the object and the PP, because the object is only one word imunyi and the following PP is also short. On the other hand, if the object is in a sentence-final position, as in

Figure 17, it is obligatorily marked by a L% boundary tone.

A H% boundary tone, on the other hand, is used to express ‘non-finality’. For example, sentences that involve an ‘if-clause’ separate the main and subordinate clauses with a H% boundary tone, followed by an obligatory pause, if the subordinate clause precedes the main clause.

This H% boundary tone as a ‘continuation rise’ is also observed in other African tone languages, such as Moro (Kordofanian, Sudan) (Rose & Piccinini, 2017), Shingazidja (Bantu, Comoros)

(Patin, 2017), Chichewa, and Tumbuka (Bantu, Malawi) (Downing, 2017). 28

One interesting feature of the H% in Ikpana is that it can occur sentence-finally in certain declarative sentences. For example, the right edge of a is marked by H% even if it appears sentence-finally. Also, when a sentence-final element is fronted to the beginning of the sentence (e.g., wh-question answers), the final syllable of the derived sentence is marked by H%.

The sentences in Figures 20 and 21 are answers to the wh-questions mɛ nu uma ɔri imunyi? ‘where does the mother hold hair?’, and mɛ nu uma ɔri imunyi bibi? ‘where does the mother hold a little hair?’, respectively. In each case, the PP elicited by the wh-questions (ɔyɔ etsi ‘under a tree’ and

ɔyɔ bibi etsi ‘under a small tree’) is fronted, and the final tone of the rest of the sentence is raised by a H% boundary tone.

Figure 20. Sample pitch track showing H% in a H-tone context when a sentence-final element (ɔyɔ etsi ‘under a tree’) is fronted.

29

Figure 21. Sample pitch track showing H% in a H-tone context when a sentence-final element (ɔyɔ bibi etsi ‘under a small tree) is fronted.

Furthermore, if a sentence involves a fronting, the final vowel of the fronted element (V1) is often preserved, deleting the first vowel of the rest of the sentence (V2), i.e., …etsi (u)m[ɔ]… in

Figures 20 and 21. One may wonder if there is a L- at the juncture between the fronted element and the rest of the sentence. The postposition etsi (‘under’), however, seems to be somehow exceptional: it is always realized with the final pitch being lower than the first pitch regardless of the position, and never affected by H%. Unfortunately, etsi is the only H-tone postposition used in sentences that involves a fronting construction. The same construction has to be tested with other

H-tone postpositions, such as otsu ‘on top of’ to determine if there is a L- after a fronted element.

As the L%, the visibility of the H% also depends on the tonal context of the final two syllables.

In Figure 20, the final two syllables show an F0 rise because the penultimate syllable is a H pitch, while in Figure 21, the final syllable is slightly higher than the penultimate syllable because the penultimate syllable is an extra-high pitch.

Finally, the lexical L-tone sentence also shows the presence of the H% boundary tone, though the effect is smaller than the lexical H-tone sentences. Figure 22 shows an L-tone declarative sentence, where the sentence-final L tone is subject to a L% boundary tone, while Figure 23 shows an answer to the wh-question mɛ nu ina ɔnɛ azayi? ‘Where does a person buy beans?’, in which the PP ɔyɔ nu (‘in a grave’) is fronted and the final syllable of the derived sentence is realized as high as the penultimate syllable due to a H% boundary tone.

30

Figure 22. Sample pitch track showing L% in a L-tone context.

Figure 23. Sample pitch track showing H% in a L-tone context when a sentence-final element (ɔyɔ nu (‘in a grave’) is fronted.

Overall, it is found that Ikpana uses L% at the edges of almost all declarative sentences, and

H% to express non-finality. However, the visibility of each boundary tone depends on the tonal context of the final two syllables. The L% lowers the final lexical tone, exhibiting a sharp fall when the penultimate syllable is raised while a flat pitch when the penultimate syllable is not raised.

In contrast, H% raises the final tone, exhibiting a flat pitch when the penultimate syllable is raised while a rise when the penultimate syllable is not raised. The phonetic realization of L% and H% is investigated in more detail in 3.3.1.3. 31

3.2.3.2 Final lengthening in the sentence-medial IP position

A sentence-medial IP is often marked by final lengthening, which can be considered as a continuation marker. Figure 24 compares the average duration of the final vowel /i/ of bibi at the sentence-final IP position and that of the same vowel at the sentence-medial IP position. The former vowel comes from the two sentences, um[ɔ] r[i] munyi bibi (uma ɔri imunyi bibi) ‘the mother holds a little hair’), and um[ɔ] ŋɔny[i] munyi bibi (uma ɔŋɔnyi imunyi bibi) ‘the mother takes a little hair by force’, while the latter vowel comes from the 11 sentences that have an additional PP after the former two sentences, e.g., {um[ɔ] r[i] munyi bibi}IP {ɔyɔ̆ etsi}IP (uma ɔri imunyi bibi ɔyɔ etsi) ‘the mother holds a little hair under a tree’. Each speaker’s data is presented in two columns in the figure. The first column (green box) for each speaker is the duration of the vowel in the sentence-final IP condition, while the second column (pink box) is that of the vowel in the sentence-medial IP condition for the same speaker. The sentence-medial IP condition for

M1 is missing because he consistently pronounced the sentences without an intervening IP boundary. As observed, the vowel is longer in the sentence-medial IP position than in the sentence- final IP position, suggesting that only sentence-medial IPs are marked by final lengthening.

32

Figure 24. Boxplot comparing the duration of the vowel at the sentence-final IP position (green box) and that of the same vowel at the sentence-medial IP position (pink box).

3.2.3.3 The IP as the domain of vowel hiatus resolution

As described in 2.2.2.1, vowel hiatus is resolved by either deletion or reduction of one of the

two vowels. In my data, deletion is more common and happens in a wide variety of contexts, while

reduction often happens when V1 is /i/ and V2 is different from /i/. Crucially, hiatus resolution is

limited by the presence or absence of an IP boundary. More specifically, hiatus resolution does not

occur when an IP boundary exists between the two vowels, but almost always occurs when there

is no intervening IP boundary, as in (4). As shown in the parentheses, some vowel hiatuses

involving a grammatical word as V1, or ones formed as a result of fronting, are resolved by a

deletion of V2. However, there are no reductions happens to V2 in my data. More systematic

analysis of vowel hiatus resolution is needed to determine in what condition which vowel is deleted.

(4) a. V → ∅ / […___ V…]IP (or […V ___…]IP)

b. V → V̆ / […___ V…]IP

The rule in (4) neatly predicts the interaction between prosodic structure and vowel hiatus

resolution. If an ip boundary occurs at a point of vowel hiatus, vowel hiatus resolution still occurs,

meaning that either vowel deletion or vowel reduction occurs. If the latter occurs, a L- tone is

realized on the reduced syllable. This is because vowel hiatus resolution in Ikpana occurs across

an ip boundary; this is different from the cases of some other languages, such as Greek (Arvaniti

& Baltazani, 2005; Baltazani, 2006) and Uyghur (Major & Mayer, 2018), whose domain of vowel

hiatus resolution is the ip boundary.

33

3.3 Other sentence types

3.3.1 Yes-no question prosody

Although it is unexpected given the typological situation that a vast majority of Kwa languages are reported to have the lax question prosody (i.e., non-high-pitched question markers),

Dorvlo (2008) states that yes-no questions in Ikpana are marked by a final H tone and final vowel lengthening. For example, a sentence-final lexical L tone is realized as a H tone on a lengthened vowel in yes-no questions.

(5) a. Statement: a-kpɛ i-va L H L L 2SG-eat thing ‘You have eaten’

b. Question: a-kpɛ i-va-a L H L H H+ 2SG-eat thing ‘Have you eaten?’ (Adapted from Dorvlo, 2008 p. 32)

However, since (5) is the only pair provided by Dorvlo to describe yes-no question prosody, it is unclear whether these markers invariably appear in any tonal contexts, such as those with a sentence-final lexical H tone. Furthermore, it is also possible that there are other markers present when various sentence structures are systematically investigated.

Accordingly, this section provides an in-depth analysis of yes-no question prosody in Ikpana, on the basis of the declarative intonation pattern described in the previous sections. It argues that

34

Ikpana yes-no questions are primarily marked by final lengthening and a H% boundary tone.

Furthermore, one of the speakers exhibits global pitch register raising.

3.3.1.1 Final lengthening

One of the primary markers of Ikpana yes-no question is final lengthening. In fact, it is quite common in the area where the lax prosody is the norm (i.e., Kwa, Gur, and Kru families), although they typically combine it with a L% boundary tone (Rialland, 2009). Figures 3 (repeated here as

Figure 25) and Figure 26 show a H-tone sentence and the corresponding yes-no question.

Compared with Figure 25, the sentence-final vowel is substantially lengthened, as shown by the double letters (i.e., bibii) in Figure 26.

Figure 25. Sample pitch track showing lack of final vowel lengthening in a declarative.

35

Figure 26. Sample pitch track showing final vowel lengthening in a yes-no question.

Figure 27 compares the duration of the sentence-final vowel in statements and that of the same vowel in the corresponding yes-no questions. The values are based on three repetitions of 36 like- tone declaratives (22 all H tone sentences and 14 all L tone sentences) and the corresponding yes- no questions. Each speaker’s data is presented in two columns. The first column (green box) for each speaker is the duration of the sentence-final vowel in the statements (S), while the second column (pink box) is that of the same vowel in the yes-no questions (Q).

Speaker Statement Question Ratio

(ms) (ms)

M1 124.95 393.84 3.15

M2 138.05 335.63 2.43

F1 149.87 490.76 3.27

Figure 27. Boxplot comparing the duration of the sentence-final vowel in statements (green box) and that of the same vowel in the corresponding yes-no questions (pink box).

The results of a mixed-effect linear regression with individual speaker as the random intercept suggest that the sentence-final vowel is significantly longer in the yes-no questions than in the statements (p < .05). Moreover, the amount of vowel lengthening is more than an addition of vocalic mora, as the average duration of the final vowel in yes-no questions is more than two times longer than in statements across all speakers, and more than three times longer than in statements in M1 and F1. 36

3.3.1.2 Register raising

Register raising is one of the most common question markers of African languages. Here the presence or absence of register raising is assessed by comparing sentence-initial F0 values between statements and yes-no questions. Figure 28 displays the average F0 values of the initial syllables of 36 like-tone statements and the corresponding yes-no questions split by lexical tone. In each graph, the first and second columns represent the L-tone values, while the third and fourth columns represent the H-tone values. Furthermore, the first and third columns represent statements (S), while the second and fourth columns represent yes-no questions (Q).

Figure 28. The average F0 values of the initial syllables of 36 like-tone statements and the corresponding yes-no questions (left: M1, middle: M2, right: F1). The values are split by lexical tone and sentence type.

A linear regression model per individual speaker was run, with the initial F0 values of the sentences as the dependent variable, while the lexical tone (H or L), sentence type (statements or questions), and the interaction between them were held as fixed factors. In addition to a significant effect of lexical tone for all speakers (p < .001), speakers M1 and M2 exhibited a marginally significant effect (p = .087) and a significant effect (p < .001) respectively of sentence type on the initial F0 values, as yes-no questions have higher initial F0 values than those in the statements, regardless of the lexical tone type. However, speaker F1 did not show any significant effect of 37

either sentence type or the interaction, indicating that F1 does not use register raising to mark questions. Interestingly, there is a complementary relationship between the effect of final lengthening and that of register raising. That is, F1, who does not exhibit register raising, lengthens final vowel more than the other two speakers, while M2, who exhibits a robust effect of register raising, shows the smallest effect of final lengthening.

Since the results suggest that there are interspeaker differences among speakers, register raising does not seem to be a robust marker of Ikpana yes-no questions. Accordingly, this study considers it to be an optional marker of yes-no questions.

3.3.1.3 H% boundary tone

Yes-no questions in Ikpana are also marked by a H% boundary tone. We will first look at H- tone sentences whose penultimate syllable is raised. Figure 26 in Section 3.3.1.1, is an example sentence, where the lexical H tone on the penultimate syllable is realized as an extra-high pitch as it is the first syllable of the adjective bibi (‘small’). As is observed, the final syllable is lengthened, and the pitch remains as high as the penultimate syllable.

Figure 29 (left) compares the average F0 values of the sentence-final H tone adjective bibi

(‘small’) in statements and yes-no questions. Each solid line represents the average F0 values of the statements for each speaker, while each dotted line represents those of the yes-no questions for each speaker. As can be observed from the figure, the final syllable is lowered in the statements, reflecting the L% boundary tone, while stays as high as the penultimate syllable in the yes-no questions, reflecting a H% boundary tone.

38

Figure 29. Left: Average F0 values of the sentence-final H tone adjective bibi (‘small’) in statements and the corresponding yes-no questions. The values are based on three repetitions of 10 statements that have ɔyɔ bibi (‘a small tree’) in the object position and the corresponding 10 yes-no questions. Right: Average F0 values of the sentence-final H tone noun ɔyɔ (‘a tree’) which follows the H-tone verb ri (‘hold’) in statements and the corresponding yes-no questions. The values are based on three repetitions of four statements that have ɔri ɔyɔ (‘hold a tree’) as their VP and the corresponding four yes-no questions.

A mixed-effect linear regression was run with the F0 value as the dependent variable, with sentence type (statement or question), position (penultimate or final), and the interaction between them as the independent variables, and with speaker as the random intercept. The results suggest that every independent variable is significant (p < .001). The significant effect of sentence type indicates an overall higher F0 in yes-no questions, while that of position indicates lower F0 in the final syllable, reflecting L% in statements. Finally, the significant interaction indicates that the effect of position

(i.e., L%) is limited to statements, suggesting that L% only appears in statements.

Figure 29 (right) shows another case where the lexical H tone on the penultimate syllable is realized as an extra-high pitch. It compares the average F0 values of the sentence-final H tone noun ɔyɔ (‘a tree’) in the VP ri ɔyɔ (‘hold a tree’) in statements and yes-no questions. The first

39

syllable of ɔyɔ is realized as an extra-high pitch because the verb H tone, which is realized with an extra-high pitch, extends to it due to vowel reduction. As in Figure 24 (left), the final syllable is lowered in the statements due to L%, while remaining flat in the yes-no questions due to H%.

Figure 30, on the other hand, shows an example pitch track of a yes-no question, where the lexical H tone on the penultimate syllable is not raised. In this case, the pitch of the final syllable is raised to the extra-high pitch level, due to the effect of H% and stays relatively flat throughout the lengthened vowel.

Figure 30. Sample pitch track showing H% and final vowel lengthening when the lexical H tone on the penultimate syllable is realized as a H pitch.

Figure 31 (left) shows the average F0 values of the sentence-final H tone noun ɔyɔ in statements and yes-no questions; however, the noun is always preceded by the L-tone verb nɛ

(‘buy’), so that the tone on the penultimate syllable is the L-verb tone, which is realized as a H pitch. In this situation, the final syllable is slightly lowered relative to the penultimate syllable in the statements, while an F0 rise is observed in the yes-no questions. Likewise, Figure 31 (right) shows the average F0 values of the last two syllables of the sentence-final H-tone noun imunyi. As in Figure 31 (left), the final syllable is slightly lower in the statements, while raised in the yes-no questions. 40

Figure 31. Left: Average F0 values of the sentence-final H tone noun ɔyɔ (‘a tree’) which follows the L-tone verb nɛ (‘buy’) in statements and the corresponding yes-no questions. The values are based on three repetitions of four sentences. Right: Average F0 values of the last two syllables of the sentence-final H tone noun imunyi (‘hair’) in statements and the corresponding yes-no questions. The values are based on three repetitions of four sentences.

The same relationship between the height of the penultimate syllable and the visibility of each boundary tone holds in L-tone sentences. Figure 32 shows an example pitch track of a yes-no question where the lexical L tone on the penultimate syllable is raised. In this case, the lexical L tone on the penultimate syllable is realized as a H pitch and the final L tone is also higher than the corresponding statement (Figure 18), due to the H%.

41

Figure 32. Sample pitch track showing H% and final vowel lengthening when the lexical L tone on the penultimate syllable is realized as a H pitch.

Figure 33 (both left and right) shows the average F0 values of the sentence-final L tone postposition nu (‘in’) and the lexical L tone on the final syllable of the preceding noun (ɔyɔ ‘a grave’ in the left figure, and vuvɔ ‘new’ in the right figure). Due to the Hω, the lexical L tone on the penultimate syllable is realized as a H pitch. In this situation, the final syllable becomes lower than the penultimate syllable in the statements, while stays relatively high in the yes-no questions.

Figure 33. Left: Average F0 values of the sentence-final L tone postposition nu (‘in’) and the preceding lexical L tone on the last syllable of ɔyɔ (‘a grave) in statements and the corresponding yes-no questions. The values are based on three repetitions of four sentences. Right: Average F0 values of the sentence-final L tone postposition nu (‘in’) and the preceding L tone on the last syllable of vuvɔ (‘new’) in statements and the corresponding yes-no questions. The values are based on three repetitions of four sentences.

Figure 34 shows an example pitch track of a yes-no question where the lexical L tone on the penultimate syllable is not raised. In this case, the lexical L tone on the penultimate syllable is realized as a L pitch and the final L tone is slightly raised in the yes-no question, due to the H%.

42

Figure 34. Sample pitch track showing H% and final vowel lengthening when the lexical L tone on the penultimate syllable is realized as a L pitch.

Figure 35 (left and right) compares the average F0 values of the sentence-final L-tone sequences, whose penultimate syllable is not raised. The left figure shows the average F0 values of the sentence-final L-tone adjective vuvɔ (‘new’), while the right figure presents these values for the last two syllables of the L-tone noun azayi (‘beans’). The pitch on the final syllable is lower than the penultimate syllable in the statements, but the pitch fall is less steeper than the case where the penultimate syllable is raised (Figure 33). In the yes-no questions, the pitch rises slightly in the male speakers’ speech, while falls slightly in the female speaker’s speech.

43

Figure 35. Left: Average F0 values of the sentence-final L tone adjective vuvɔ (‘new’) in statements and the corresponding yes-no questions. The values are based on three repetitions of eight sentences. Right: Average F0 values of the last two syllables of the sentence-final L tone noun azayi (‘beans’) in statements and the corresponding yes-no questions. The values are based on three repetitions of four sentences.

Finally, Figure 36 (left) shows the speaker-normalized values (z-score) of the sentence-final

H tone sequences, while Figure 36 (right) shows those of the sentence-final L tone sequences. In both figures, red lines represent the cases where the penultimate syllable is raised (/H/ → [extra- high], /L/ → [H]), and the blue lines represent those where the penultimate syllable is not raised

(/H/ → [H], /L/ → [L]). As in the previous figures, solid lines represent statements and dotted lines represent yes-no questions.

Although the final pitch is lower when the penultimate syllable is an extra-high pitch (red solid line) than when it is a H pitch (blue solid line) in the left figure, it would be safe to say that there are roughly two targets on the final syllable for each lexical tone, namely lower target for

L% and higher target for H%. However, since both lexical tones are influenced by the boundary tones. the distinction between the lexical H and L tones are maintained sentence-finally, representing a case of accommodation in the typology proposed by Hyman and Monaka (2011).

Moreover, there are two phonetic patterns worth mentioning here. First, the F0 of the penultimate syllable is always higher in the yes-no questions than in the statements. This can be because of the effect of a boundary tone or that of the global register raising. Second, as argued in Section 3.2.1, the L-sequences tend to fall more than the H-tone sequences, exhibiting a greater effect of declination.

44

Figure 36. Left: Speaker-normalized values (z-score) of the sentence-final H tone sequences in statements and the corresponding yes-no questions. The values are based on three repetitions of the 22 sentences included in Figures 27 and 29. Right: Speaker-normalized values of the sentence- final L tone sequences in statements and the corresponding yes-no questions. The values are based on three repetitions of the 20 sentences included in Figures 31 and 33.

3.3.2 Wh-questions

Wh-questions in Ikpana are characterized by the use of wh-words, which may be optionally moved to sentence-initial position (wh-movement), as in (a), or remain in their base-generated positions (wh-in-situ), as in (b).

(6) a. mɛ u-ma ɔ-ri?

what CM-the mother SM-hold ‘What does the mother hold?’ b. u-ma ɔ-ri mɛ?

CM-the mother SM-hold what ‘What does the mother hold?’

45

The wh-word mɛ (‘what’) is always realized with an extra-high pitch, which can be considered as either a Hω at the end of a monosyllabic word or an idiosyncratic feature of the wh-word since it is common in other languages that wh-word is prosodically similar to a focused word. As in yes- no questions, both wh-in-situ and wh-movement questions are marked by a H% boundary tone, but no final lengthening is observed.

3.3.2.1 Intonation in wh-in-situ questions

As in yes-no questions, wh-in-situ questions are marked by a H% boundary tone, whose effect is observable when the wh-word is not in sentence-final position. Figure 37 shows a pitch track of a wh-in situ question, where the wh-word mɛ (‘what’) is located in the object position, followed by the PP ɔyɔ nu (‘in a grave’). In the sentence, the vowel /ɛ/ in the wh-word is deleted and the following vowel /ɔ/ is re-syllabified with the onset of the wh-word, i.e., m[ɔ]. The extra-high pitch is realized on the re-syllabified vowel, but the lexical L tone of the initial vowel of ɔyɔ is also realized, resulting in a falling tone. Figure 38 shows the answer to the question shown in Figure

37. The sentence-final L tone is slightly raised by H% in the wh-in-situ question in Figure 37, but lowered by L% in the answer in Figure 38.

Figure 37. Sample pitch track showing H% marking the end of a wh-in-situ question.

46

Figure 38. Sample pitch track showing L% marking the end of a wh-question answer.

Although there is no pause delimiting the wh-word, the pitch of the syllable preceding the wh- word is lowered in Figure 37. There might be a prosodic boundary before the wh-word, but it would be necessary to examine more data to confirm this claim.

Furthermore, the answer to the wh-question (i.e., azayi ‘beans’ in Figure 38) is not usually marked by any prosodic means of highlighting the information, such as higher pitch or stronger intensity. Interestingly, Kügler and Genzel (2011) found that in Akan both H and L tones are realized significantly lower under corrective focus construction. Future study should quantitatively investigate focus prosody in Ikpana.

3.3.2.2 Intonation in wh-movement questions

As in wh-in-situ questions, wh-movement questions are marked by a H% boundary tone.

Figure 39 shows a pitch track of a wh-movement question in a successive H tone context, where the final H tone is higher than the penultimate H tone.

47

Figure 39. Sample pitch track showing H% marking the end of a wh-movement question, in which the penultimate syllable is realized with a H pitch.

Figure 40 shows a pitch track of another wh-movement question, in which the penultimate syllable is realized as an extra-high pitch. In this tonal context, the sentence-final tone is as high as, or even slightly higher than the penultimate syllable.

Figure 40. Sample pitch track showing H% marking the end of a wh-movement question, in which the penultimate syllable is realized with an extra-high pitch.

In natural speech, there is no pause after the wh-word, and even the initial vowel (class marker) of the subject is deleted due to vowel hiatus resolution.

48

3.3.3 Imperatives

Imperatives are formed by using a verb stem without a subject. Figure 41 shows both the citation form of the H-tone verb phrase r[i] munyi (‘hold hair’) (left) and the imperative form r[i] munyi! (‘hold hair!’) (right). Although no tonal change is observed, the imperative form is realized with an overall higher pitch and a stronger intensity (shown as darker formants in the spectrogram) than the corresponding citation form.

Figure 41. Sample pitch track and spectrogram showing the higher pitch and stronger intensity of a H-tone imperative sentence.

Figure 42 shows the L-tone verb phrase n[a] zayi (‘buy beans’) (left) and the imperative form n[a] zayi! (‘Buy beans!’). As in Figure 41, the imperative form is realized with an overall higher pitch and stronger intensity.

49

Figure 42. Sample pitch track and spectrogram showing the higher pitch and stronger intensity of an L-tone imperative sentence.

4. Conclusion

This study investigated the types of prosodic units that exist in Ikpana, how each prosodic unit is marked by intonation, and how these units interact with the domain of segmental phonological rules. In Section 3.2, it was shown that Ikpana has a PW, an ip, and an IP, and that only the IP is the domain of vowel hiatus resolution. A PW is marked by a High boundary tone, labeled Hω, an ip is marked by the Low boundary tone, L-, and pitch reset, and an IP is marked by a boundary tone, i.e., either L% or H%, followed by a pause. A sentence-medial IP boundary is also marked by final lengthening as a continuation marker. Moreover, this study also uncovers the effect of each boundary tone on each lexical tone. The L% lowers the final lexical tone, exhibiting a sharp fall when the penultimate syllable is raised, but a flat pitch when the penultimate syllable is not raised. In contrast, H% raises the final tone, exhibiting a flat pitch when the penultimate syllable is raised, but a rise when the penultimate syllable is not raised. Crucially, both lexical tones are influenced by the boundary tones, maintaining the distinction between them (i.e., accommodation in Hyman and Monaka, 2011)

Furthermore, the model neatly predicts how these prosodic markers interact with each other.

When multiple prosodic units coincide, the prosodic markers of a larger unit delete those of any smaller unit. More specifically, the effect of the Hω is overridden by that of L- in the ip-final position, and the effect of L- is overridden by that of the IP-final boundary tone (L% or H%) in the IP-final position. Also, if an ip-boundary occurs at a point of vowel hiatus, the effect of L- is observed on the reduced syllable. This is because vowel hiatus resolution occurs across an ip boundary. 50

Finally, in Section 3.3, the intonation of other sentence types was analyzed. This section revealed that yes-no questions are marked by final lengthening, a H% boundary tone, and an optional register raising, wh-questions are marked by a H% boundary tone, and imperatives can be marked by overall higher pitch and stronger intensity. As discussed in the introduction, the Kwa family is known for its lax question prosody. Thus, it is unexpected that Ikpana uses a H% boundary tone, which is one of the tense question markers. Since Ikpana combines a H% boundary tone (a tense question marker) with final lengthening (a lax question marker), its question prosody is characterized as a hybrid form. This information may be useful not only for an understanding of the typology of question prosody, but also for the linguistic classification of GTM languages, especially because Tutrugbu, another GTM language, is also reported to exhibit an F0 rise at the end of yes-no questions (Essegbey, in press). Furthermore, wh-questions are also marked by a H% boundary tone, which bucks the cross-linguistic tendency that wh-questions have a similar intonational profile to statements (Bolinger, 1978; Downing & Rialland, 2017).

There are several limitations in this study. First, the number of lexical items used for each lexical category is quite small. It is necessary to use multiple lexical items even for a single syntactic construction to confirm that the observed pitch movement is not due to an idiosyncratic feature of the specific item, but due to postlexical F0 variation in Ikpana. Particularly, more adjectives, subject nouns and verbs have to be checked to determine the nature of the Hω. Second, the number of speakers analyzed in this study is only three. Although they are quite similar in terms of basic prosodic features, there are some interpersonal differences observed (e.g., register raising). Data from more speakers should be quantitatively examined to back up the claims put forward in this study. Finally, little is described on focus prosody. To establish a more

51

comprehensive model of Ikpana intonational phonology, focus prosody should be examined by a carefully designed experiment.

52

References

Ameka, F. K. (2017) The Ghana-Togo Mountain languages: Introduction. Language Typology and

Universals 70. Berlin and Boston: de Gruyter Mouton, 239-244.

Ameka, F. K. & Essegbey, J. (2017). Divergence and convergence among the Ghana-Togo

Mountain languages. Introduction. Language Typology and Universals 70. Berlin and Boston:

de Gruyter Mouton, 245-271.

Arvaniti, A. & Baltazani, M. (2005). Intonational Analysis and Prosodic Annotation of Greek

Spoken Corpora. In S.-A. Jun, Prosodic Typology: The Phonology of Intonation and Phrasing.

Oxford: Oxford University Press, 84-117.

Baltazani, M. (2006). Focusing, Prosodic Phrasing, and Hiatus resolution in Greek. In D. Goldstein,

D. Whalen, & C. Best, (eds), Laboratory Phonology 8. Berlin: Mouton de Gruyter, 473-494.

Beckman, M. & Pierrehumbert, J. (1986). Intonational structure in Japanese and English.

Phonology Yearbook, vol. 3, 225-309.

Blench, R. (2006). A comparative study of the Ghana-Togo Mountain languages. Cambridge.

(Ms.)

Blench, R. (2009). Do the Ghana-Togo Mountain languages constitute a genetic group? Journal

of West African Languages 36, 19-35.

Boersma, P. & Weenink, D. (2017). Praat: doing phonetics by computer [Computer program].

Version 6.0.31.

Bolinger, D. (1978). Intonation across languages. In J. Greenberg (ed.), Universals of Human

Language. Stanford: Stanford University Press, 371-425.

53

Cahill, M. (2015). Polar Question Intonation in Five Ghanaian Languages. In Selected Proceedings

of the 44th Annual Conference on African Linguistics, ed. R. Kramer, E. C. Zsiga, & O. T.

Boyer, 28-36. Somerville, MA: Cascadilla Proceedings Project.

Casali, R. F. (1996). Resolving Hiatus. PhD dissertation, University of California, Los Angeles.

Casali, R. (1997). Vowel elision in hiatus contexts: Which vowel goes? Language 73. 493-533.

Clements, G. N. & Rialland, A. (2008) as a phonological area. In B. Heine & D. Nurse

(eds.), A Linguistic Geography of Africa. Cambridge: Cambridge University Press, 36-85.

Connell, B. (2003). Four tones and downtrend: a preliminary report on pitch realization in Mambila.

In: Kotey, P. (ed), New Dimensions in African Linguistics and Languages. Trends in African

Linguistics, 3. Trenton, N.J.: Africa World Press, 75-88.

Connell, B. (2017). Tone and Intonation in Mambila. In L. J. Downing & A. Rialland, Intonation

in African tone languages. Berlin and Boston: de Gruyter Mouton, 131-166.

Connell, B. & Ladd, D. R. (1990). Aspects of pitch realization in Yoruba. Phonology 7: 1-29.

Cruttenden, A. (1997). Intonation. 2nd edition. Cambridge: Cambridge University Press.

Dorvlo, K. (2008). A grammar of Logba (Ikpana). PhD dissertation, Leiden University.

Downing, L. J. (2017). Tone and intonation in Chichewa and Tumbuka. In L. J. Downing & A.

Rialland, Intonation in African tone languages. Berlin and Boston: de Gruyter Mouton, 365-

391.

Downing, L. J. & Rialland, A. (eds), (2017). Intonation in African tone languages. Berlin and

Boston: de Gruyter Mouton.

Egbokhare, O. F. (1990). Vowel elision in Emai. Afrika Und Ubrsee 73, 19-35.

Essegbey, J. (in press). Tutrugbu (Nyangbo) language and culture. Leiden: Brill.

Genzel, S. (2013). Lexical and post-lexical tones in Akan. Potsdam: PhD dissertation Universität

54

Potsdam.

Godjevac, S. (2005). Transcribing Serbo-Croatian Intonation. In S.-A. Jun, Prosodic Typology:

The Phonology of Intonation and Phrasing. Oxford: Oxford University Press, 146-171.

Hyman, L. & Monaka, K. C. (2011). Tonal and non-tonal intonation in Shekgalagari. In S. Frota,

G. Elordieta & P. Prieto (eds), Prosodic categories: production, perception and

comprehension, 267-290. Dordrecht: Springer Verlag.

Ipek, C. (2015). The phonology and phonetics of Turkish intonation. PhD dissertation, University

of Southern California.

Jun, S.-A. (1993). The phonetics and phonology of Korean prosody. PhD dissertation, Ohio State

University.

Jun, S.-A. (1998). The Accentual Phrase in the Korean prosodic hierarchy. Phonology 15, 189-

226.

Jun, S.-A. (ed), (2005). Prosodic Typology: The Phonology of Intonation and Phrasing. Oxford:

Oxford University Press.

Jun, S.-A. (ed), (2014). Prosodic Typology II: The Phonology of Intonation and Phrasing. Oxford:

Oxford University Press.

Jun, S-A. & Fletcher J. (2014). Methodology of studying intonation: from data collection to data

analysis. In S.-A. Jun, Prosodic Typology II: The Phonology of Intonation and Phrasing.

Oxford: Oxford University Press, 493-519.

Kügler, F. (2017). Tone and intonation in Akan. In L. J. Downing & A. Rialland, Intonation in

African tone languages. Berlin and Boston: de Gruyter Mouton, 89-129.

Kügler, F. & Genzel, S. (2011). On the prosodic expression of pragmatic prominence: the case of

pitch register lowering in Akan. Language and Speech 55, 331-359.

55

Ladd, D. R. (1996/2008). Intonational Phonology, Cambridge: Cambridge University Press.

Laniran, Y. O. (1992). Intonation in tone languages: the phonetic implementation of tones in

Yoruba. Ph.D. dissertation, Cornell University.

Laniran, Y. O. & Clements, G. N. (2003). Downstep and high raising: interacting factors in Yoruba

tone production. Journal of Phonetics 31: 203-250.

Leben, W. R. & Ahoua F. (1997). Prosodic domains in Baule. Phonology 14. Cambridge:

Cambridge University Press, 113-132.

Lindau, M. (1986). Testing a model of intonation in a tone language. Journal of the Acoustical

Society of America, 80, 757-764.

Major, T. & Mayer, C. (2018). Towards a phonological model of Uyghur intonation. Proc. 9th

International Conference on Speech Prosody Posnań, 744-748.

Mudzingwa, C. (2010). Shona morphophonemics: Repair strategies in Karanga and Zezuru. PhD

dissertation, University of British Columbia.

Nespor, M. & Vogel, I. (1986). Prosodic phonology. Dordrecht: Foris.

Ohala, J. (1984). An ethological perspective on common cross-language utilization of F0 in .

Phonetica 41, 1-16.

Patin, C. (2017). Tone and intonation in Shingazidja. In L. J. Downing & A. Rialland, Intonation

in African tone languages. Berlin and Boston: de Gruyter Mouton, 285-320.

Pierrehumbert, J. (1980). The phonology and phonetics of English intonation. PhD dissertation,

Massachusetts Institute of Technology.

Pierrehumbert, J. & Beckman, M. (1988). Japanese tone structure. Cambridge, Mass.: MIT Press.

Pulleyblank, D. (1988). Vocalic underspecification in Yoruba. Linguistic Inquiry 25, 344-353.

56

Rialland, A. (2007). Question prosody: an African perspective. In C. Gussenhoven and T. Riad

(eds.), Tones and tunes: studies in word and sentence prosody, 35-62. Berlin: Mouton de

Gruyter.

Rialland, A. (2009). The African lax question prosody: its realisation and geographical distribution.

Lingua 199. 928-949.

Rose, S. & Piccinini, P. (2017). Intonation in the Thetogovela dialect of Moro. In L. J. Downing

& A. Rialland, Intonation in African tone languages. Berlin and Boston: de Gruyter Mouton,

19-50.

Selkirk, E. (1984). Phonology and syntax: the relation between sound and structure. Cambridge,

Mass.: MIT Press.

Selkirk, E. (1986). On derived domains in sentence phonology. Phonology Yearbook 3. 371-405.

Truckenbrodt, H. (1999). On the relation between syntactic phrases and phonological phrases.

Linguistic Inquiry 30. 219-282.

Urua, E. A. (2002). The tone system of Ibibio. In U. Gut, & D. Gibbon, (eds), Typology of African

Prosodic System. Bielefeld: University of Bielefeld, 65-85.

Westerman, D. (1930). A study of the . Oxford: Oxford University Press.

Yip, M. (2002). Tone. Cambridge: Cambridge University Press.

57